AI Transfer via the Ontological Invariant

Modern artificial intelligence, including large language models, stores knowledge in weights, leading to catastrophic forgetting and architecture lock-in. Current transfer methods (fine-tuning, LoRA, distillation) fail because they copy data—numerical weight values—rather than structural information. This paper proposes an applied methodology based on the Ontological Invariant (OI) framework to transfer the structural pattern, designated as the mind invariant (K_mind), between neural network architectures. In this context, K_mind is conceptualized not as a set of weights, but as a topological graph of activations, such as connections between attention heads in a Transformer or state transitions in an SSM. The success of the transfer is verified through a structure identifier (Id) using the criterion Id(K_mind) = Id(K′_mind) on the new carrier, even across different architectures. To operationalize this approach, the paper introduces a five-stage Structural Transfer Procedure, version 1 (PP1), detailing each stage from extraction and encoding to deployment and adaptation. Crucially, PP1 integrates control mechanisms at every step, using Id calculations and functional metrics to localize errors and ensure structural integrity. This approach aligns with Class A (Productive Bridge) of the MBM series, providing a conceptual link between ontological theory and machine learning engineering. As a pilot paper in the OIE series, it establishes a conceptual methodology without experimental validation. Open questions, including the strict mathematical formalization of Id for activation patterns and the operational metrics for complexity gradients, are outlined for future research in subsequent series publications.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24285562
Primary Topic
Domain Adaptation and Few-Shot Learning
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

AI Transfer via the Ontological Invariant

Alexander Yourievitch Kotelnikov
Zenodo (CERN European Organization for Nuclear Research)
Domain Adaptation and Few-Shot Learning
preprint

AI Transfer via the Ontological Invariant

Alexander Yourievitch Kotelnikov
preprint en

Abstract

Modern artificial intelligence, including large language models, stores knowledge in weights, leading to catastrophic forgetting and architecture lock-in. Current transfer methods (fine-tuning, LoRA, distillation) fail because they copy data—numerical weight values—rather than structural information. This paper proposes an applied methodology based on the Ontological Invariant (OI) framework to transfer the structural pattern, designated as the mind invariant (K_mind), between neural network architectures. In this context, K_mind is conceptualized not as a set of weights, but as a topological graph of activations, such as connections between attention heads in a Transformer or state transitions in an SSM. The success of the transfer is verified through a structure identifier (Id) using the criterion Id(K_mind) = Id(K′_mind) on the new carrier, even across different architectures. To operationalize this approach, the paper introduces a five-stage Structural Transfer Procedure, version 1 (PP1), detailing each stage from extraction and encoding to deployment and adaptation. Crucially, PP1 integrates control mechanisms at every step, using Id calculations and functional metrics to localize errors and ensure structural integrity. This approach aligns with Class A (Productive Bridge) of the MBM series, providing a conceptual link between ontological theory and machine learning engineering. As a pilot paper in the OIE series, it establishes a conceptual methodology without experimental validation. Open questions, including the strict mathematical formalization of Id for activation patterns and the operational metrics for complexity gradients, are outlined for future research in subsequent series publications.

Zenodo (CERN European Organization for Nuclear Research)
Domain Adaptation and Few-Shot Learning
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon